Medical image reconstruction method and system

Abstract

The invention relates to a medical image reconstruction method, which comprises the following steps of S1, acquiring CT image data of a patient with a neurodegenerative disease, and preprocessing the CT image data; and S2, a Control-Net branch is newly added on the basis of a Stable Diffusion model, and image reconstruction is carried out on the preprocessed image data by utilizing the Stable Diffusion and the Control-Net so as to generate a high-quality PET (Positron Emission Tomography) image. The invention also relates to a medical image reconstruction system. According to the method, specific prior information is added in the image reconstruction process, and the precision and reliability of image reconstruction are improved.

Description

Technical Field

[0001] The invention relates to a medical image reconstruction method and system. Background Art

[0002] Parkinson's disease (PD) is the fastest growing neurological disease in the world, driven by an aging population. Currently, there are no blood or laboratory tests for non-genetic cases of Parkinson's disease, and clinical diagnosis mainly relies on patient history and neurological examination. The disease has become an increasingly severe public health challenge, especially in many countries with scarce medical resources, where patients have difficulty obtaining timely care and drug support. As the disease progresses, imaging changes are often difficult to detect in the early stages, so the development of early quantitative diagnostic methods has become key to diagnosis and treatment.

[0003] Existing medical imaging technologies such as CT, MRI and PET provide information on anatomical structure, soft tissue details and metabolic activity respectively. However, a single imaging mode is difficult to provide complete disease information. Therefore, multimodal image fusion has become an important research direction, especially the combination of CT and PET images. In medical image processing, traditional image reconstruction methods usually rely on manual analysis, which has certain subjectivity and errors. With the rise of computer-aided diagnosis (CAD) technology and deep learning, quantitative analysis can provide more accurate image reconstruction and disease diagnosis.

[0004] Currently, image reconstruction and conversion technologies based on deep learning have been applied in the medical field. For example, the image-to-image translation method is used to convert CT images into PET images, which can combine anatomical information with metabolic activities. However, most of these technologies rely on traditional convolutional neural networks (CNNs), which still face the problems of insufficient accuracy and low computational efficiency when dealing with complex scenes and details.

[0005] In summary, the prior art has at least the following disadvantages:

[0006] First, lack of accuracy in early diagnosis: existing imaging technologies have difficulty accurately detecting early lesions of Parkinson's disease, especially tiny structural changes.

[0007] Second, the efficiency of image reconstruction is low: traditional image reconstruction methods rely on manual intervention, are inefficient and easily affected by subjective factors.

[0008] Third, insufficient technical integration: Although multimodal image fusion has great advantages in theory, in practical applications, there are still many challenges in how to effectively fuse CT and PET images. Summary of the invention

[0009] In view of this, it is necessary to provide a medical image reconstruction method and system.

[0010] The present invention provides a medical image reconstruction method, which comprises the following steps: S1, collecting CT image data of patients with neurodegenerative diseases and performing preprocessing; S2, adding a Control-Net branch on the basis of a Stable Diffusion model, and performing image reconstruction using Stable Diffusion and Control-Net for the preprocessed image data to generate high-quality PET images.

[0011] Preferably, the method further comprises:

[0012] Step S3, performing quantitative analysis on the generated PET image to extract features of the lesion area.

[0013] Preferably, the method further comprises:

[0014] Step S4, using SPM12 to extract the values ​​of each brain region to evaluate its performance in terms of diagnostic accuracy and reliability.

[0015] Preferably, the step S1 comprises:

[0016] CT image data of Parkinson's disease patients were collected, and image registration and standardization were performed to ensure that the data met the input requirements of the Stable Diffusion model.

[0017] Preferably, the step S2 comprises:

[0018] A new Control-Net branch was added based on the Stable Diffusion model;

[0019] All noise points, prompt words and additional control images are input into Control-Net and output to the Stable Diffusion model after processing;

[0020] The Stable Diffusion model combines the prompt words to generate a PET image controlled by CT;

[0021] Image reconstruction using Stable Diffusion and Control-Net.

[0022] Preferably, the Control-Net is an innovative neural network architecture designed for the diffusion model, which achieves precise control of the generation process by introducing additional conditional inputs; and divides the network structure into two parts: a trainable part and a non-trainable part.

[0023] Preferably, the step S3 comprises:

[0024] The reconstructed PET images are quantitatively analyzed to extract the characteristics of the lesion area; the characteristics of the lesion area include: changes in brain regions and metabolic activities.

[0025] The present invention provides a medical image reconstruction system, which includes a preprocessing module and an image reconstruction module, wherein:

[0026] The preprocessing module is used to collect CT image data of patients with neurodegenerative diseases and perform preprocessing;

[0027] The image reconstruction module is used to add a Control-Net branch based on the Stable Diffusion model, and use Stable Diffusion and Control-Net to reconstruct the preprocessed image data to generate high-quality PET images.

[0028] Preferably, the system further comprises a quantitative analysis module:

[0029] The quantitative analysis module is used to perform quantitative analysis on the generated PET image to extract the characteristics of the lesion area.

[0030] Preferably, the system further comprises an evaluation module:

[0031] The evaluation module is used to extract the values ​​of each brain region using SPM12 to evaluate its performance in terms of diagnostic accuracy and reliability.

[0032] The present invention utilizes deep learning technology to reconstruct and fuse CT and PET images of Parkinson's patients to achieve efficient and accurate detection of lesion areas. The Stable Diffusion model is used to generate high-quality medical images, especially the reconstruction of PET images. The generation process is finely controlled in combination with Control-Net to ensure that the generated image is consistent with the original CT image in anatomical structure. Control-Net provides conditional input in this process to ensure that specific prior information is added during the image reconstruction process to improve the accuracy and reliability of the reconstruction. Image feature extraction and quantitative analysis are performed in combination with a deep learning model to provide accurate support for the early diagnosis of Parkinson's disease. The present invention has the following advantages:

[0033] First, efficient and accurate image reconstruction: The combination of Stable Diffusion and Control-Net can generate high-quality, low-noise images, improving the accuracy and detail of the images.

[0034] Second, the accuracy of early diagnosis is improved: quantitative analysis combined with deep learning technology can identify the lesion area earlier and more accurately, especially in the early stages of Parkinson's disease.

[0035] Third, automated processing: The technical solution of the present invention greatly reduces manual intervention and improves diagnostic efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a flow chart of the medical image reconstruction method of the present invention;

[0037] Figure 2 A schematic diagram of an overall model provided by an embodiment of the present invention;

[0038] Figure 3 A schematic diagram of the effect of reconstructing a high-quality PET image provided by an embodiment of the present invention;

[0039] Figure 4 FIG. 4 is a hardware architecture diagram of the medical image reconstruction system of the present invention. DETAILED DESCRIPTION

[0040] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0041] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0042] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] Example 1

[0044] See also Figure 1FIG. 1 is a flowchart of a preferred embodiment of the medical image reconstruction method of the present invention.

[0045] This embodiment is described using Parkinson's disease as an example, and can also be extended to other types of neurodegenerative diseases, such as Alzheimer's disease, amyotrophic lateral sclerosis, etc.

[0046] Please also read Figure 2 , step S1, collecting CT image data of Parkinson's disease patients and performing preprocessing. Specifically:

[0047] CT image data of Parkinson's disease patients were collected, and image registration and standardization were performed to ensure that the data met the input requirements of the Stable Diffusion model.

[0048] Step S2: Add a Control-Net branch based on the Stable Diffusion model, and use Stable Diffusion and Control-Net to reconstruct the preprocessed image data to generate high-quality PET images. Figure 3 Specifically:

[0049] This embodiment adds a Control-Net branch based on the Stable Diffusion (SD) model. All noise points, prompt words and additional control images are input into Control-Net, and then output to the SD main control model after processing. Finally, the SD main control model combines the prompt words to generate a PET image controlled by CT.

[0050] Image reconstruction using Stable Diffusion and Control-Net: The Stable Diffusion model is used to generate high-quality PET images, and Control-Net is used to ensure the structural consistency of the generated images with the original CT images. This process introduces prior conditions so that the generated images are not only consistent with the original CT in terms of functional information, but also have accurate anatomical structures.

[0051] Control-Net is an innovative neural network architecture designed for diffusion models. It achieves precise control of the generation process by introducing additional conditional inputs. Its core feature is that the network structure is divided into two parts: ① The trainable part is used to learn the deep representation of the control variables; ② The untrainable part (locked) retains the original characteristics of the stable diffusion model. This design ensures that the model maintains the generation ability and stability of the original diffusion model while introducing conditional constraints, thereby effectively improving the generation effect in complex scenarios.

[0052] Step S3, quantitatively analyze the generated PET image to extract the characteristics of the lesion area. Specifically:

[0053] The reconstructed PET images are quantitatively analyzed to extract the features of the lesion area, and early diagnosis is performed in combination with a deep learning algorithm. The features of the lesion area include changes in brain regions and metabolic activities.

[0054] Conduct a Stable Diffusion image reconstruction model based on Control-Net: Optimize the application of Control-Net in the Stable Diffusion framework, reconstruct medical images of Parkinson's disease, and enhance feature extraction, especially the early changes in the lesion area.

[0055] Step S4, use SPM12 to extract the values ​​of each brain region to evaluate its performance in terms of diagnostic accuracy and reliability. Specifically:

[0056] By combining the high-quality PET images generated by Stable Diffusion with deep learning technology, an early diagnosis model for Parkinson's disease was constructed, and SPM12 was used to extract numerical values ​​of various brain regions to evaluate its performance in diagnostic accuracy and reliability.

[0057] For further information, see Figure 2 In this example, a Control-Net branch is added based on Stable Diffusion (SD). All noise points, prompt words and additional control images are input into Control-Net and output to the SD main control model after processing. Finally, the SD main control model combines the prompt words to generate a PET image under CT control. Figure 3 The two branches are independent of each other, and there is no need to retrain a completely new model. At the same time, Control-Net can also flexibly reference and process control images.

[0058] Example 2

[0059] This embodiment is described using Parkinson's disease as an example, and can also be extended to other types of neurodegenerative diseases, such as Alzheimer's disease, amyotrophic lateral sclerosis, etc.

[0060] See also Figure 4 , is a hardware architecture diagram of the medical image reconstruction system 10 of the present invention. The system comprises: a preprocessing module 101, an image reconstruction module 102, a quantitative analysis module 103, and an evaluation module 104. Among them:

[0061] The preprocessing module 101 is used to collect CT image data of Parkinson's disease patients and perform preprocessing. Specifically:

[0062] The preprocessing module 101 collects CT image data of Parkinson's disease patients and performs image registration and standardization processing to ensure that the data meets the input requirements of the Stable Diffusion model.

[0063] The image reconstruction module 102 is used to add a Control-Net branch based on the Stable Diffusion model, and use Stable Diffusion and Control-Net to reconstruct the image data after preprocessing to generate high-quality PET images. Figure 3 Specifically:

[0064] This embodiment adds a Control-Net branch based on the Stable Diffusion (SD) model. All noise points, prompt words and additional control images are input into Control-Net, and then output to the SD main control model after processing. Finally, the SD main control model combines the prompt words to generate a PET image controlled by CT.

[0065] Image reconstruction using Stable Diffusion and Control-Net: The Stable Diffusion model is used to generate high-quality PET images, and Control-Net is used to ensure the structural consistency of the generated images with the original CT images. This process introduces prior conditions so that the generated images are not only consistent with the original CT in terms of functional information, but also have accurate anatomical structures.

[0066] Control-Net is an innovative neural network architecture designed for diffusion models. It achieves precise control of the generation process by introducing additional conditional inputs. Its core feature is that the network structure is divided into two parts: ① The trainable part is used to learn the deep representation of the control variables; ② The untrainable part (locked) retains the original characteristics of the stable diffusion model. This design ensures that the model maintains the generation ability and stability of the original diffusion model while introducing conditional constraints, thereby effectively improving the generation effect in complex scenarios.

[0067] The quantitative analysis module 103 is used to perform quantitative analysis on the generated PET image to extract the characteristics of the lesion area. Specifically:

[0068] The quantitative analysis module 103 performs quantitative analysis on the reconstructed PET image, extracts the characteristics of the lesion area, and performs early diagnosis in combination with a deep learning algorithm. The characteristics of the lesion area include: changes in brain regions and metabolic activities.

[0069] The quantitative analysis module 103 performs a Stable Diffusion image reconstruction model based on Control-Net: optimizes the application of Control-Net in the Stable Diffusion framework, reconstructs medical images of Parkinson's disease, and strengthens feature extraction, especially early changes in the lesion area.

[0070] The evaluation module 104 is used to extract the values ​​of each brain region using SPM12 to evaluate its performance in terms of diagnostic accuracy and reliability. Specifically:

[0071] By combining the high-quality PET images generated by Stable Diffusion with deep learning technology, an early diagnosis model for Parkinson's disease was constructed, and SPM12 was used to extract numerical values ​​of various brain regions to evaluate its performance in diagnostic accuracy and reliability.

[0072] For further information, see Figure 2 In this example, a Control-Net branch is added based on Stable Diffusion (SD). All noise points, prompt words and additional control images are input into Control-Net and output to the SD main control model after processing. Finally, the SD main control model combines the prompt words to generate a PET image under CT control. Figure 3 The two branches are independent of each other, and there is no need to retrain a completely new model. At the same time, Control-Net can also flexibly reference and process control images.

[0073] In addition to the early diagnosis of neurological diseases, the present invention can also be applied to other medical imaging fields that require fine image reconstruction and quantitative analysis, such as tumor detection and diagnosis of cardiovascular diseases.

[0074] The present invention combines the Stable Diffusion model with the Control-Net deep learning framework to provide new possibilities in the field of medical image processing. The Stable Diffusion model integrated with Control-Net is used to improve the reconstruction effect of Parkinson's disease medical images and accurately present the early lesion characteristics. In particular, in the early quantitative diagnosis of the disease, the accuracy and efficiency of diagnosis can be significantly improved.

[0075] Although the present invention has been described with reference to the current preferred embodiments, those skilled in the art should understand that the above-mentioned preferred embodiments are only used to illustrate the present invention and are not used to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the scope of protection of the present invention.

Claims

1. A medical image reconstruction method, characterized in that: The method comprises the following steps: S1, collect CT image data of patients with neurodegenerative diseases and perform preprocessing; S2, a Control-Net branch is added based on the Stable Diffusion model, and Stable Diffusion and Control-Net are used to reconstruct the preprocessed image data to generate high-quality PET images.

2. The method according to claim 1, characterized in that The method further comprises: Step S3, performing quantitative analysis on the generated PET image to extract features of the lesion area.

3. The method according to claim 2, characterized in that The method further comprises: Step S4, using SPM12 to extract the values ​​of each brain region to evaluate its performance in terms of diagnostic accuracy and reliability.

4. The method according to claim 3, characterized in that: The step S1 comprises: CT image data of Parkinson's disease patients were collected, and image registration and standardization were performed to ensure that the data met the input requirements of the Stable Diffusion model.

5. The method according to claim 4, characterized in that The step S2 comprises: A new Control-Net branch was added based on the Stable Diffusion model; All noise points, prompt words and additional control images are input into Control-Net and output to the StableDiffusion model after processing; The Stable Diffusion model combines the prompt words to generate a PET image under CT control; Image reconstruction using Stable Diffusion and Control-Net.

6. The method according to claim 5, characterized in that The Control-Net is an innovative neural network architecture designed for the diffusion model, which achieves precise control of the generation process by introducing additional conditional inputs; the network structure is divided into two parts: a trainable part and a non-trainable part.

7. The method according to claim 6, characterized in that The step S3 comprises: The reconstructed PET images are quantitatively analyzed to extract the characteristics of the lesion area; the characteristics of the lesion area include: changes in brain regions and metabolic activities.

8. A medical image reconstruction system, characterized in that: The system includes a preprocessing module and an image reconstruction module, wherein: The preprocessing module is used to collect CT image data of patients with neurodegenerative diseases and perform preprocessing; The image reconstruction module is used to add a Control-Net branch based on the Stable Diffusion model, and use Stable Diffusion and Control-Net to reconstruct the preprocessed image data to generate high-quality PET images.

9. The system according to claim 8, characterized in that The system also includes a quantitative analysis module: The quantitative analysis module is used to perform quantitative analysis on the generated PET image to extract the characteristics of the lesion area.

10. The system according to claim 9, characterized in that The system further comprises an evaluation module: The evaluation module is used to extract the values ​​of each brain region using SPM12 to evaluate its performance in terms of diagnostic accuracy and reliability.

Images